Some literature reports how learners' alternative ideas in science may be coherent, stable and theory-like. However, other commentators suggest that the available data supports the view that children's thinking is inconsistent, with elicited notions being piecemeal, ad hoc and deeply situated in specific contexts. This is considered to reflect the fragmentary and unscientific nature of the learner's knowledge. Accumulating evidence from in-depth work with individual learners is beginning to show that models of cognitive structure that can usefully inform teaching may need to be more complex than either of these views admit. Evidence from a case study is presented to show how a learner may simultaneously hold several alternative explanatory schemes, each of which is persistent over time and applied coherently across a wide range of overlapping contexts. It is argued that the manifold nature of learners' conceptions may be a key to modelling conceptual development.